基于Jeston TK1的并行计算系统设计与实现

Design and implementation of parallel computing system based on Jeston TK1

针对当前分布式计算集群的功耗高、体积大、携带不方便的问题,设计实现一种小型化、高性能的嵌入式并行计算系统。首先,以多台高性能嵌入式平台Jeston TK1为硬件基础,裁剪专业的嵌入式系统作为支撑系统;在详细分析Jeston TK1的性能和Hadoop分布式架构的基础上,构建Hadoop分布式系统;然后,采用MapReduce并行计算模式,提出Jeston TK1平台下Hadoop架构的并行计算框架,并设计并行计算流程和实现过程;最后,采用Java语言编程实现并行计算程序,通过具体的实验案例对系统进行分析和性能验证。在实例计算实验中,设计的并行计算系统能稳定地运行并行计算任务,并且在相同输入的情况下,计算时间由单机的146 s和伪分布式系统的105 s减少到23 s。实验结果表明,设计的嵌入式并行计算系统具有高效的计算性能,可以实现英文单词的快速检索和统计。

Concerning the problems of high power consumption,large volume and inconvenient portability of distributed computing cluster,a miniaturized and high performance embedded parallel computing system was designed and implemented.Firstly,based on the hardware of Jeston TK1,which is a multi-high performance embedded platform,a professional embedded system was clipped as the supporting system.On the basis of detailed analysis of the performance of TK1 and the distributed architecture of Hadoop,a Hadoop distributed system was constructed.Then,using MapReduce parallel computing mode,a parallel computing framework under Jeston TK1 platform was proposed,and the parallel computing process and implementation process were designed.Finally,the parallel computing program was implemented by Java language programming,and the system was analyzed and verified through specific experimental cases.In the case of computing experiment,the designed parallel computing system can run parallel computing tasks stably,and the computing time was reduced from 146 seconds by a single computer and 106 seconds by a pseudo-distributed system to 23 seconds under the same input.Experimental results show that the designed embedded parallel computing system has high efficient computing performance,which can realize fast retrieval and statistics of English words.